Three-dimensional Reconstruction Based On Images

Three-dimensional reconstruction is an important research area in computer vision. In reality, 3D reconstruction is widely used in many fileds, such as object recognition, industrial automated design, video games, animation, creation of digital heritage archive, and restoring the original appearance of buildings, etc. At the same time, in order to achieve a real-time computing, it also demands good efficiency in reconstruction.In this paper, we do the reconstruction based on the images taken from various perspectives of the real objects. We do the research of reconstruction mainly in camera calibration, model generation and optimization, and texture mapping .Ultimately the original model of the object is reconstructed with the image sequences. The results of the reconstruction is good, while it is also has good feasibility.We take the calibration methods based on calibration object whose shape is known, using a series of mathematical calculations we get the camera parameters, while calibrating the camera with the tool developed by Klaus Strobl.We take the the silhouette-based as the main reconstruction algorithm. With the silhoutte of the object getting through the calibration results ,We computer the boundingbox and then get the visual hull by doing the intersection test to the bounding box,and then subdivising it with Marching Cubes algorithm. Then, preliminary modeling completed.The early model includes much noise,and is very rough,so it needs to be optimized. It includes mesh smoothing and mesh simplification. We use the Laplacian-based smoothing algorithm in smoothing part,and side-delete algorithm in simplification part while it makes a minimum error. The result is good.In texture mapping, the improved model plane parametrization is applied, then generated the correlation between the model and the texture. We use it to get all the texture infomation of pixels related the model, and finally, we get a new texture. With the OpenGL function, we finished the texture mapping.According to this research, the reconstruction technology based on images is realistic. Finally, we introduce the application of the reconstruction results in the laboratory platform, demonstrating scalable applications of reconstruction.